Deep-Penetrating Uncrewed Aerial Vehicle-Based Ground-Penetrating Radar for Landslide Surveys: A framework for high-resolution migration imaging and bedrock interface extraction

Landslides pose significant risks to lives and infrastructure, and their evolution depends critically on the morphology of the bedrock interface. Unmanned aerial vehicle (UAV)-based ground-penetrating radar (GPR) offers a promising noncontact geophysical solution for mapping these complex structures. However, in complex mountainous terrains, the dual challenges of dense vegetation and steep undulating topography often degrade data quality, leading to severe signal attenuation and imaging artifacts. Leveraging a self-designed low-frequency UAV-based GPR system, we develop a high-precision algorithmic framework tailored for deep-penetrating landslide surveys. To mitigate structural ambiguities induced by rugged topography, we first propose reverse-time migration (RTM) formulated in a curvilinear coordinate system. An orthogonal body-fitted grid (OBFG) is utilized to eliminate staircase approximation errors, ensuring high-fidelity wavefield focusing even under steep topographic conditions. To automate the interpretation of the enhanced RTM images, we introduce a probabilistic dynamic programming (DP)-based technique that extracts the bedrock interface along with its 80% probabilistic uncertainty band from the migration profiles. To validate the robustness of this framework, field surveys are conducted over two types of geologically distinct scenarios in Sichuan Province, China. Furthermore, the effects of transmission power and flight height on data quality and imaging performance are systematically evaluated. Our results indicate that the proposed UAV-based GPR system successfully penetrates to depths reaching 20 m in complex terrains. Quantitative validation against borehole data shows that the absolute errors are below 1 m, and the relative errors are below 10%. This article provides effective geophysical insights and a robust technical workflow for the early identification and structural characterization of landslide hazards in mountainous regions.